Highlights:

  • Taalas’ custom chip funding was provided by Quiet Capital and Pierre Lamond, a prominent venture capitalist known for backing several early players in the semiconductor industry, with roots tracing back to Fairchild Semiconductor and National Semiconductor in the 1960s.
  • According to The Information, Taalas is developing processors designed to meet the needs of specific neural networks, going beyond mere AI optimization.

Taalas Inc., a startup specializing in custom chips for AI models, secures USD 50 million to bolster its commercialization endeavors.

Taalas’ custom chip funding was provided by Quiet Capital and Pierre Lamond, a prominent venture capitalist known for backing several early players in the semiconductor industry, with roots tracing back to Fairchild Semiconductor and National Semiconductor in the 1960s.

Taalas is under the Chief Executive Officer of Ljubisa Bajic, who founded the AI chip startup Tenstorrent Inc. in 2016 prior to joining Taalas. The valuation of the latter company following its most recent $200 million funding round was $1 billion. Lejla Bajic, Drago Ignjatovic, and Bajic, who both held engineering leadership positions at Tenstorrent, co-founded Taalas in August of last year.

Most AI chips are optimized to accelerate matrix multiplications, key mathematical operations utilized by neural networks for processing data. Certain processors incorporate extra optimizations tailored for particular AI applications. Nvidia Corp.’s latest H200 graphics card, for instance, comes equipped with ample high-speed memory to enhance the performance of language models.

Taalas intends to expand upon the notion. As reported by The Information, the organization is developing processors that will be constructed with the specifications of a particular neural network in consideration, as opposed to merely being optimized for AI. With this strategy, the company anticipates that its processors will be considerably quicker than the graphics devices of today.

Bajic stated, “Commoditizing AI requires a 1000x improvement in computational power and efficiency, a goal that is unattainable via the current incremental approaches. The path forward is to realize that we should not be simulating intelligence on general purpose computers, but casting intelligence directly into silicon.”

Creating a bespoke processor can require years and hundreds of millions of dollars in investment in certain instances. Consequently, Taalas’ strategy to develop multiple chips, each optimized for a distinct AI algorithm, is expected to pose significant technical hurdles. To overcome these challenges, the company is creating an automated engineering workflow to expedite its semiconductor design processes.

According to Taalas, one of its processors will have sufficient RAM to accommodate an “entire large AI model.” Executing a model solely on-chip eliminates the necessity for external RAM, reducing data travel between the RAM module and the chip.
Minimizing data travel, in turn, accelerates processing.

Taalas’ custom chips aims to complete the tape-out of its inaugural product, a processor tailored for large language models, in the third quarter.  A tape-out refers to the final version of a chip design sent to the fabrication facility for manufacturing. Taalas aims to begin shipping the processor to customers in early 2025.